I've done spec-driven development with Claude Code for almost all of the past year. The agent writes a spec, I approve it, and the agent builds it. The specs run beyond thousands of words, and I started approving them after skimming one or two. I have ADHD, so a wall of agent prose loses me fast.
decipher rewrites an agent-written spec, PR, report, or reply for a reader who lost the thread. You get a gist of a few words, a bottom line, and one question you answer to yourself. Then come the decisions and risks that are yours, and the steps in order. It also tries to warn you, in one line, when a doc dodges what you asked or builds on a guess.
I used to use i-have-adhd, caveman, etc, but none of them would give me what I needed. It's been some weeks since I've been using decipher, and recently open-sourced it.
decipher rewrites an agent-written spec, PR, report, or reply for a reader who lost the thread. You get a gist of a few words, a bottom line, and one question you answer to yourself. Then come the decisions and risks that are yours, and the steps in order. It also tries to warn you, in one line, when a doc dodges what you asked or builds on a guess.
I used to use i-have-adhd, caveman, etc, but none of them would give me what I needed. It's been some weeks since I've been using decipher, and recently open-sourced it.
Install in Claude Code:
It also works in Codex, OpenCode, and Cursor.It's free and MIT-licensed. Eager to hear your feedback!